Google NotebookLM File Size Limit: Source Caps and Large-Corpus Storage

Google's official Gemini Notebook documentation establishes a hard per-file ceiling of 200MB and 500,000 words per source, capped at 50 sources per notebook on the free plan as of September 2026. The file size limit for Gemini Notebook (formerly NotebookLM) restricts individual uploaded sources to a maximum of 200MB or 500,000 words across up to 50 sources per notebook.

Google rebranded NotebookLM to Gemini Notebook in July 2026 to unify its research assistant under the primary Google AI umbrella, while preserving the underlying source-grounded architecture. In September 2026, Google introduced compute-based usage limits that meter chat queries, deep research reports, and multimedia generation based on prompt complexity and session length. Despite these interface and brand updates, the core file ingestion constraints remain firm.

Here is the exact breakdown of limits governing individual sources and notebook containers:

• Per-File File Size: 200MB maximum for local file uploads across all document types.
• Per-File Word Count: 500,000 words per individual document, article, or video transcript.
• Total Source Count: 50 sources per notebook on the free Standard plan, scaling up to 600 on premium tiers.
• Total Notebooks: 100 notebooks per user account on the free plan, expanding to 500 on paid plans.
• Supported Formats: PDF (.pdf), Microsoft Word (.docx), Plain Text (.txt), Markdown (.md), CSV (.csv), PowerPoint (.pptx), ePub (.epub), web page URLs, public YouTube URLs with captions, local audio files, and Google Drive files.

These parameters define two separate boundaries. The first is container capacity: how many discrete items you can attach to one notebook. The second is item density: how large and text-heavy an individual file can be before the ingestion pipeline rejects it. Understanding where your collection collides with these boundaries determines whether you can work within Gemini Notebook directly or need an external storage architecture.

In May 2026, Google aligned NotebookLM quotas with its unified Google AI subscription tiers. Upgrading to a paid tier expands the number of sources you can attach to a notebook and increases your daily query quotas, but it leaves the per-source file size and word caps untouched.

The following comparison details the operational limits across every plan tier, verified against Google's documentation:

The critical takeaway from this structure is that paid subscriptions solve volume problems, not individual file size problems. Upgrading from the free plan to Google AI Pro multiplies your source allowance from 50 to 300 items, and Google AI Ultra pushes the ceiling to 600 items. However, if you possess an archival PDF that measures over 200MB or an uncompressed export containing over 500,000 words, no subscription tier will allow you to upload it directly.

Collaboration rules introduce another operational constraint. In Google's official Gemini Notebook documentation, sharing a notebook does not change the source limit for any collaborator. When an Ultra user creates a notebook containing 550 sources and shares it with a colleague on the free plan, the collaborator can view and query all 550 sources. However, free collaborators cannot add new sources beyond the notebook's existing allocation, and they remain subject to their personal daily chat query caps.

Direct-upload AI tools treat document storage as an attachment mechanism. You upload files into a closed container, the vendor runs an extraction process, and an LLM reads across the assembled text. While convenient for reviewing several reports or a single textbook, this pattern breaks when applied to enterprise research, legal discovery, or multi-gigabyte technical archives.

The first point of failure is physical scan bloat. Legal filings, municipal records, historical manuscripts, and architectural submittals frequently exist as scanned image files. Even with modern compression, a lengthy deposition binder or municipal building code scan easily exceeds maximum upload allowances. Because direct-upload tools enforce an arbitrary per-file upload gate, teams spend hours manually segmenting documents into arbitrary volumes simply to bypass upload validation.

The second failure is format fragmentation. Enterprise projects do not consist solely of clean 10-page Word documents. They include technical whitepapers, CSV exports of sensor logs, CAD reference tables, and presentation decks. Uploading these individually rapidly exhausts the 50-source limit on free accounts, forcing researchers to merge unrelated files into bloated mega-documents.

The third failure is repository isolation. In Gemini Notebook, each notebook exists as an isolated silo. You cannot execute a query that spans Notebook A (Patent Filings) and Notebook B (Engineering Specifications). To cross-reference findings, you must duplicate source files into a third notebook, consuming duplicate source slots and fracturing version control. When an original file updates in Google Drive, synchronized updates reflect only in notebooks where that specific file link was established.

Move Beyond NotebookLM File Size Limits with Intelligent Storage

Store large research collections in an intelligent workspace with hybrid semantic search and versioning built in. Connect any assistant via MCP without hitting file size limits. Every organization starts with a 14-day free trial.

Start 14-Day Trial

Solving the file limit problem does not require waiting for AI vendors to increase their upload caps. The professional architecture decouples document persistence and indexing from the AI chat interface. Instead of pushing whole files into Gemini Notebook or Claude Projects, teams keep their primary corpus in an intelligent cloud workspace like Fast.io Workspaces, where files are indexed automatically upon arrival, and connect their preferred assistant via the Model Context Protocol (MCP).

This workflow operates through three coordinated layers:

1. Centralized Workspace Ingestion

Teams store their multi-gigabyte research collections in shared, organization-owned Fast.io workspaces. Files can be uploaded directly using chunked uploads, which process multi-gigabyte files without browser timeouts or file size ceilings. Alternatively, teams can use cloud import to pull existing repositories from Google Drive, Dropbox, Box, or OneDrive. Google Drive imports today with sync coming soon, while Dropbox, Box, and OneDrive support synchronization on a schedule or on demand.

2. Intelligence Mode and Hybrid Indexing

Once files land in the workspace, enabling Intelligence Mode activates automatic document indexing for retrieval-augmented generation (RAG). Fast.io indexes PDFs, text files, spreadsheets, and presentations for hybrid search, combining full-text keyword matching with semantic vector search. The workspace maintains an indexed knowledge layer without requiring teams to configure external vector databases, write chunking scripts, or manage embedding pipelines. When queries execute, the intelligence layer returns direct answers backed by document citations that link to exact source passages.

3. Remote MCP Integration

Rather than uploading documents into each assistant, you connect your assistant directly to your workspace using Fast.io's remote Model Context Protocol endpoint at https://mcp.fast.io/mcp. When your AI assistant needs information, it queries the workspace index using MCP tools, retrieves the relevant excerpts, and incorporates only those specific passages into its context window. This architecture does not alter or raise Google's or Anthropic's vendor-specific upload limits; instead, it bypasses the need to upload files into those containers entirely.

{
  "mcpServers": {
    "fastio": {
      "url": "https://mcp.fast.io/mcp/key",
      "headers": {
        "Authorization": "Bearer YOUR_FASTIO_API_KEY"
      }
    }
  }
}

If your immediate workflow requires using Gemini Notebook directly, several practical preprocessing techniques can help you compress and restructure documents to clear the 200MB and 500,000-word ceilings.

1. Segment Large PDFs with CLI Utilities

When a comprehensive legal record, municipal filing, or technical handbook exceeds the single-file limit, segmenting the document into logical chapters or topical appendices allows you to upload each section as an individual source.

On macOS and Linux, you can split large PDFs without quality loss using qpdf:

qpdf input-large-corpus.pdf --pages . 1-150 -- chapter-1.pdf
qpdf input-large-corpus.pdf --pages . 151-300 -- chapter-2.pdf

2. Downsample Raster Graphics in Scanned Documents

Oversized PDFs usually stem from uncompressed bitmap scans rather than actual text volume. If a document consists of scanned pages, downsampling the embedded raster images to 150 DPI provides sharp, readable text for Gemini Notebook's optical character recognition while cutting file size to a fraction of the original uncompressed scan.

You can reprocess scanned PDFs using Ghostscript:

gs -sDEVICE=pdfwrite -dCompatibilityLevel=1.4 -dPDFSETTINGS=/ebook \
   -dNOPAUSE -dQUIET -dBATCH \
   -sOutputFile=compressed-output.pdf input-oversized-scan.pdf

The setting downsamples internal raster images, which comfortably pulls oversized scans below upload ceilings without stripping textual fidelity.

3. Strip Media Streams and Upload Text Transcripts

Uploading raw MP3 or M4A audio files directly to Gemini Notebook expends file bandwidth and introduces transcription latency. For long meetings, hearings, or video lectures, extract the audio track to a plain text transcript locally using open-source tools like Whisper before uploading. A lengthy recorded deposition that occupies hundreds of megabytes as raw audio compresses down to a lightweight plain text transcript, importing into Gemini Notebook in seconds while consuming a modest portion of your word ceiling.

4. Extract Tabular Data into Structured Metadata Views

Large CSV datasets or tabular appendices often breach the 500,000-word limit while offering poor analytical results inside conversational LLM prompts. For document collections containing hundreds of invoices, receipts, lease agreements, or structured forms, use Metadata Views inside your workspace.

Metadata Views transform unstructured files into a typed, queryable database. You describe the target fields in plain English, and the system automatically designs a schema spanning Text, Integer, Decimal, Boolean, URL, JSON, and Date & Time types. Connected assistants can then query structured values, filter by date ranges, and calculate totals via MCP without consuming document upload slots or overloading model context windows.